Settings pages

The Kiki desktop app exposes preferences in the Settings dialog and activity on separate pages such as Usage in the sidebar. This page helps you find each control and what it changes; the linked pages hold the full reference.

The settings sidebar groups twenty entries under seven headings: App, Models & agents, Sessions & work, Capabilities, Permissions & rules, Connection & remote, and Advanced. Four of them merge what used to be separate pages and are switched by tab — Models & providers, Agents, Extensions & tools, and Browser & computer — so a path below that ends in a tab name is one click inside the entry.

"I want to change X" quick lookup:

I want to change…	Go to
Sign in with a subscription, or add an API key	Models & agents → Models & providers → Providers & accounts
How outbound requests identify themselves	Models & agents → Models & providers → Identity
Version and update channel (Stable / Beta)	Advanced → About & updates
Agent profiles, prompt-field overrides	Models & agents → Agents → Agent profiles
Low-frequency long-term memory reminders	loop_control.continuity_cadence.memory_maintenance in config.toml
Whether input drafts are remembered	App → General
Move selected settings to another home or device	App → General → Settings transfer
Background-task limits and scheduled tasks	Sessions & work → Sessions
Search and retrieval configuration source	Capabilities → Search & retrieval
Installed plugins, skills, and MCP servers	Capabilities → Extensions & tools
Subagent dispatch configuration	Models & agents → Agents → Subagents & dispatch for server-wide defaults; Child agents on a single profile remains in Agent profiles
Run a browser for agent tasks	Capabilities → Browser & computer → Browser control
Drive the server's own desktop	Capabilities → Browser & computer → Computer control
Experimental features	Advanced → Developer & labs, under Where experimental features live
Inspect this session's prompt sources	Session header ⋯ → Effective prompts
Temporarily use the preview space for the cockpit	Right-panel header Cockpit
Token usage and cost	Usage → History
Running or queued model requests	Usage → Live
Model/provider request concurrency	Usage → Live → Concurrency limits
Send usage to another service	Usage → External sync
TUI theme, editor, and other CLI-side preferences	The /config commands in the terminal, see CLI counterpart

Import and export settings

Open Settings → App → General → Settings transfer, then Import or export. Export lets you select several categories, with item counts: models and providers, agents and prompts, permissions and sessions, tools/MCP/plugins, browser and computer, notifications/IM, and appearance/device preferences. Include credentials is off by default. Turn it on only when the downloaded JSON file will stay private; the file contains the stored credentials, not masked placeholders. External account sign-ins are not transferred.

On the target, choose Import and select a kiki.settings JSON file up to 8 MiB. Check the target home and space, select the categories, then review the new, conflicting and skipped counts. Keep existing, the default, preserves existing IDs, keys and resources, including explicit inherited restrictions. Replace selected replaces only keys and resources present in the selected categories, retaining everything else. Importing without credentials also retains the target's existing credentials. Cancel writes nothing.

The package contains explicit settings from the source home, valid user Profiles and SYSTEM.md, referenced prompt TOML files, user MCP definitions, and saved device preferences. It does not copy inherited defaults, project or system-wide profile directories, chat history, caches, installed plugin/Recipe packages, or skin assets. Install required components separately and check machine-specific paths and commands on the target. A missing or unsafe referenced prompt file prevents the package from being exported or applied.

Include credentials also transfers search API keys saved in Kiki, checked against their provider, slot and endpoint bindings. Environment-variable names in settings are retained, but keys from process environment variables or another app's login or credential files are not copied. MCP definitions with inline header or token arguments are omitted when credentials are not selected.

After Apply, Kiki reads the saved values back and reports each changed item's status. A partial result means some items were saved; review the statuses and preview again before retrying. Configuration is reloaded after saving. Profile changes and settings that apply only to new sessions take effect in a new session; an existing conversation keeps its bound prompt sources until you rebuild its context.

Connections

Settings → Models & agents → Models & providers opens Providers & accounts (/settings/ai?tab=providers), the default tab; Defaults (/settings/ai?tab=defaults) and Identity (/settings/ai?tab=identity) sit beside it. /settings/providers and /settings/models open the same tab. The page has three blocks, in this order, plus a folded Maintenance section at the bottom.

① API key providers lists hosted APIs you reach with a key. An empty list in this block does not mean the page has no connections.

Add connection adds a service you reach with a key. With nothing connected yet, the form is the page; otherwise it opens beside the lists. It asks which service, and how:

• From the directory — pick a service from models.dev and Kiki fills in its address and model list. An id that already exists updates that connection, and the form says so before you press.
• Enter it myself — a protocol, an address and a key, for a service the directory does not have, including a local server.

For either path, use Add model inside the connection form, then choose Save provider. Once any connection exists, search covers every configured connection and also finds a connection by the IDs and names of its models. The new row appears in the block for how it is reached.

② Account sign-in (subscription) lists the subscription accounts already connected: Kimi Code, a ChatGPT account for Codex, a Grok Build account, GitHub Copilot. An account that is not connected yet has Sign in in this block. An account you already have a connection for is not offered again; a sign-in that has gone stale is recovered on that connection's own row.

③ Local HTTP and CLI holds a server on this machine and local programs such as AGY CLI, Codex, and Claude Code. Once a CLI program is enabled, choose it from the conversation's runtime picker. Let an outside client use Kiki opens Connection & remote → Access from elsewhere → Agents & clients (/settings/connection-services?tab=clients). /settings/external-clients opens the same tab. The Browser tab holds web access and the Cloudflare tunnel; Other Kiki holds the Kikis allowed to connect to you.

An account connection is written by the sign-in, so its row has no protocol, address or key to fill in. Expand it to see which account is behind it, what the vendor says it has left, and the action that changes the sign-in:

• Connected — the credential works, and Kiki renews it on its own. Sign out ends that sign-in inside Kiki and removes the models it added; at the provider you stay signed in.
• Sign-in expired — the provider no longer accepts it. Sign in again replaces it on this same connection; nothing is added to the list.
• Sign-in didn't finish — the flow ended without a credential: you declined it, the code expired, or it failed. One line says which, and the row is left as the server reports it.
• Waiting for you — a sign-in is running, with the code, Open verification page, how long it stays valid, and Cancel sign-in.

A completed sign-in adds that account's models to the catalog on the server. In Providers & accounts, expand the connection row to see its Models list and sign-in state.

Signing out here affects Kiki only. The provider keeps its own record of your subscription, and signing in again reconnects the same account.

Reusing a sign-in this machine already has

For ChatGPT (Codex) and Grok Build, a connection can use the sign-in their own app already holds on this machine instead of a new one. Kiki reuses it and renews it when it runs out; it does not copy the credential, does not start the other app, and signing in or out here does not change anything there.

"This machine" is the machine Kiki's server runs on. That is usually the one you are looking at; when it is not, Look somewhere else on the server takes a directory to read from instead.

Check this machine reports which account is on the other side and where it is kept — a file, the system keyring, or an encrypted store — and Use this sign-in then attaches that exact account. A credential that was replaced in between is refused rather than adopted, and one that is due for renewal is used and renewed.

If the machine cannot be used, the page says why in terms you can act on — no account for that service, signed out, stored somewhere unreadable, or belonging to a different account — and offers nothing to attach.

Stop using it lets go of the machine's sign-in for this connection. It removes Kiki's reference and the models it provisioned; the sign-in itself is still there, and the other app is unaffected.

Timeouts

Open Settings → Connection & remote → Server connection, then Current connection and HTTP requests (/settings/connection?tab=current). These settings govern requests from this interface to the Kiki server, not the provider connections above:

• Request timeout (seconds) — the default deadline for ordinary requests this interface makes to the Kiki server.
• Reading timeout (seconds) — how long a history or content read may wait before Kiki gives up on it. The default is 0, meaning no deadline: a large or slow conversation loads as long as it takes. Set a number of seconds if you would rather a stalled read fail quickly.

A reading timeout only bounds waiting; it never limits how much can be read. A changed value applies to the next read — it does not interrupt one already running, and it does not reconnect anything. Leave it at 0 unless you specifically want a ceiling. These are GUI settings, not keys in config.toml; the server-side memory budgets that govern how much history is kept in memory are separate — see transcript_memory.

Models and Recipes

Settings → Models & agents → Models & providers opens Providers & accounts. To edit a model, expand its connection row, open its Models list, and click the model; the full editor opens in a side panel. Defaults (/settings/ai?tab=defaults) is where the other task models are set. Identity (/settings/ai?tab=identity) decides how outbound requests introduce themselves. /settings/ai?tab=models and /settings/models open Providers & accounts, while /settings/ai?tab=providers&model=<model id> opens that model's editor directly.

Above the tabs, the New sessions use bar and the star on a model row set the global model for new sessions. The titles, fast model, and subagents listed beside it open Defaults, where you change them.

In the model editor, change request settings and prompt text, then Save once. The identity selector keeps Shared, Main agent, and Delegated agent drafts separate; saving includes edits made in all three. Advanced fields edit the same model rather than a second form. Editing text loaded from a file stores an inline copy on the model, leaving the source file unchanged. Remove model is at the bottom of the editor and asks for confirmation.

Opening a prompt does not by itself leave the shared group. While Main agent or Delegated agent still follows that group, the first Save of a prompt copies the whole effective group onto that identity, including unchanged file references and timing. Later edits to the shared group no longer change it. A steering source set to Follow the user-message setting uses the steering text and timing already in effect for user messages on the selected identity. That is the saved steering, not the words of the user message. After the identity has its own prompt copy, this reads the steering on that copy. A source left unconfigured stays off.

Usage fields on an identity are separate from that copy. Changing Default thinking effort, Service tier, Max generated tokens, Automatic compaction point, or Context budget writes only that field. When a shared Context budget or Max generated tokens is already set, a higher value on the identity does not raise it, and a lower value tightens it. Default thinking effort, Service tier, and Automatic compaction point take the identity's value. Use shared again restores that field; leaving a field out of the save keeps it.

Model switching is under Defaults on that page (/settings/ai?tab=defaults). It sets the default and Ask before switching. Direct keeps the current context, Summarize first compacts it, and Fresh context starts a new window. Exception rules match canonical model ids. * and ? are the only wildcards, an empty side matches any model, and the first enabled rule wins. Save rules replaces that list and does not change the default or the ask setting. The fields are [model_switch].

A Recipe contributes a reusable set of prompt and model-setting overrides. In the model editor, choose Choose a Recipe, then browse installed packages, configured markets or Import. Preview a URL or an absolute path on the server; HTTPS ZIP imports also need their SHA-256 checksum. Reading or selecting a package does not apply it.

• Apply to model commits the selected Recipe together with the model's current unsaved edits. Restore manual settings removes that model's Recipe reference and saves those edits too; it does not erase saved manual values or a profile's separate Recipe.
• Install and apply installs the preview and performs that same model save. Install only stores the package without applying it or saving model drafts. If applying fails, the installation, selection and drafts remain; correct the reported problem and retry Apply to model without reinstalling.
• Copy and edit creates an independent local Recipe; Extend from this keeps a parent reference. In its author view, edit Prompt words or Source files, then Save Recipe. This saves the package, not a new model selection. Mixed inline/file text stays in separate editable segments. A notice appears when a prompt edit will rewrite the manifest's TOML formatting; file-only edits leave that manifest unchanged.
• Download as ZIP shares the accepted package, including inherited content, without uploading it. Save author drafts first. A profile can also reference an installed Recipe through its Markdown frontmatter.

Recipe values cover only the fields or prompt slots they declare; uncovered settings keep their lower-layer values, and explicit profile differences still apply. A running conversation keeps its bound revision. Use Rebuild context while idle to adopt current sources.

The Identity tab is the central page for request identity: the global default, the built-in and custom identities, client versions, where each identity is used, and recent requests. A provider or a single model carries its own identity in that editor's Request identity section, and the later layer wins. An old /settings/identity address opens this tab. See Request identity.

About

Settings → Advanced → About & updates (/settings/about) shows the current version and owns the update settings. Update channel picks Stable or Beta, and every check — automatic or manual — reads the channel from here. Check for updates automatically turns the daily check on or off; with it off, only Check for updates checks. The When an update is found setting applies: Notify me shows a dialog with the version and a short summary and waits for you, while Download and install starts the install without that dialog. Either way, Kiki asks before closing spaces that still have running work.

The dialog's three choices are saved rather than repeated each time: Remind me tomorrow comes back 24 hours later, Skip this version silences that version on that channel only, and a newer version still comes either way. Before installing, Kiki re-checks the channel and the version and refuses an offer that has changed underneath you. If a choice cannot be saved, the dialog stays open and asks you to try again. See Kiki desktop.

Memory

Settings → Memory (/settings/memory) separates the overall Allow memory switch from Memory Advisor. The advisor card lets you choose a configured model and protocol, confirm a destination for each workspace, and test the connection with sample data. Its Advanced fold contains selection and request budgets. The memory-page header and Ctrl-K provide the same global advisor switch; turning it off retains the model and restores the previous memory selection. See Memory Advisor for setup and what is sent.

Agents

Settings → Models & agents → Agents has two tabs. Agent profiles (/settings/agents?tab=profiles), the default, lists the selected workspace's main and subagent profiles in one table — role, engine, model, effort, and who may dispatch whom — and marks the file each profile comes from. Open a row to edit its instructions and settings; a file-backed profile can also be edited at the source shown. The same tab holds Identity and agent loading and Prompt fields, which edits the [prompt] prompt-field overrides section in config.toml; expand the card first. Subagents & dispatch (/settings/agents?tab=dispatch) holds the defaults that apply across profiles — see Dispatch capabilities. See Agents and subagents and Prompt field overrides.

The periodic long-term memory reminders run on their own schedule. To stop just those while keeping the new standing-instruction and pre-compaction checks, set memory_maintenance = false in config.toml; memory tools, approval and task notes stay available. See Continuity reminder settings.

Notifications & messages

Settings → App → Notifications & messages (/settings/notifications) is its own entry in the settings sidebar. It has three tabs: Channels & accounts (/settings/notifications?tab=accounts), When to notify (/settings/notifications?tab=rules), and This device (/settings/notifications?tab=device). Old ?tab=im and ?tab=channels links open Channels & accounts; ?tab=away opens This device.

Channels & accounts has an IM accounts & channels card for two-way IM accounts such as QQ, Telegram, and WeChat. Its Push channels card lists one-way bots and webhooks and provides Choose a built-in channel. QQ's Configure account link opens the kiki-qq plugin settings. QQ 0.4.0 offers mobile QQ QR setup alongside manual credentials and OneBot settings. Confirm the bot on your phone, check the returned AppID, then choose Save this account. Saving leaves a stopped connection stopped; an enabled connection reconnects using the new credentials.

When to notify has Notification rules: the master Send notifications switch, quiet hours, Skip the session you are viewing, and Skip short work. Its Which channel hears what table has one row per push channel and one checkbox column for each event, The conversation finishes and A question is waiting; The conversation finishes uses the same work-completion decision. This table is the only place to choose which events a channel receives; per-channel event checkboxes are no longer in each channel row. Previously disabled masters, connections and channels stay disabled. New configurations default to on with Skip short work at 0 seconds, so short replies qualify too. Raising it filters short-work notices for those channels, without changing what completion means. Channel rules are saved by the connected server.

Under This device, System notifications, Allow system notifications controls session notifications and pending-input reminders from other spaces on this device. The default-on Notify when the conversation finishes switch notifies when you are not viewing that conversation, including when another conversation or Settings is open in Kiki. Completion means the assigned work has settled: the main agent has finished, finite background tasks and subagents have settled, and their results and any automatic continuation have been handled. A turn ending alone does not notify. Resident servers, watchers and future schedules do not keep work unfinished; approvals, questions, failures and user stops are not completion notices. In a browser, allow this site to send notifications; if you previously denied permission, restore it in the address bar’s site permissions.

Open the notified conversation to clear its unread badge. Focusing Kiki, visiting Home or Settings, or opening another conversation does not mark it read. Notification clicks open the same conversation in its original home and connection; other conversations remain unread. A conversation left open in a hidden or unfocused window is not marked read until you return to it. After a brief disconnect or app reopen, Kiki catches up completion receipts retained by the running server for ten minutes, without repeating receipts already observed on this device. The first launch does not replay historical completions. System-notification switches are saved on this device.

Search & retrieval

Settings → Capabilities → Search & retrieval opens Overview & source (/settings/search). An old /settings/environment-tools address opens this page, and a tab there selects a search tab (overview, search, fetch, providers, or advanced). It inspects the built-in search and retrieval module — the capability behind the WebSearch and FetchURL tools — showing which configuration source is in effect and whether the server reuses the local search configuration. The equivalent configuration lives in config.toml — see Configuration files.

The same section's Advanced & diagnostics tab carries a Full-text index card for searching your own session history. The first index is built in the background, newest sessions first, and the card counts what is done so far. If the index stops for a reason you can fix, that card is where Restart the indexer appears; when indexing is off or unavailable by configuration, there is nothing to restart and the reason is stated instead.

Browser control

Settings → Capabilities → Browser & computer opens on its Browser control tab (/settings/browser-control?tab=browser), which brings together the overall browser verdict, the connection list, and four setup sources. Computer control is the second tab.

Allow browser sits at the top of the page beside a single verdict line that states whether the agent can currently use a browser, how many browsers are allowed, which default is active, or the first gate that blocks browser control. It stays allowed unless this server has turned it off. On by default chooses whether a new conversation starts with the tools. A conversation can turn them off, or back on while they are allowed. The next model request uses that choice. The allow switch does not change which browsers are installed. It makes browser tools available subject to the conversation’s tool settings. If the open conversation still has an older tool selection, Apply current tool settings previews the target conversation and updates only its tools after confirmation. Its prompt, model and history stay unchanged; explicit tool selections and denied tools remain restricted. Older bindings whose tool source is unknown are never widened automatically.

Under Browser connections, every saved browser appears in one list with its Allow agent use checkbox, connection type, location, runtime state, and a read-only Default tag when it is the active default. Check a browser's box to let the agent use it; clear it to take the permission back immediately. Default connection on the list header is the single place to choose which browser new sessions use; Choose each time stores no default. Allowing a browser and choosing the default are separate: clearing a checkbox leaves the default where it is (and the verdict line calls out when the selected default is not allowed), and a failed default change leaves the allowed browsers as they were so you can try again.

Pressing + Add browser opens the four browser sources Kiki can prepare or check, followed by Configure manually; the same four sources also stay manageable in the Browser sources fold under the connection list:

• Kimi Browser Extension: drives the browser you already use, with your own sign-ins. Kiki installs the plugin and local bridge; add the extension from your browser's store and approve it there. An existing usable bridge can connect even when its version differs from the bundled download. Advanced options → Check and update downloads the verified release; a failed update keeps the previous installation, and a running bridge stays on its current version. The WebBridge readiness fold inside this route states whether the local bridge is working, offers Check again, and — while a component is missing — the consented Set up WebBridge action.
• Independent browser: Kiki manages a separate browser and sign-in profile. Set up downloads its components; Connect starts control, and Open browser opens or activates its window. Advanced options provides Check and update and, on a Windows server, Create desktop shortcut. Updates keep a running browser unchanged; shortcuts appear on the server's desktop without overwriting unrelated shortcuts.
• Codex / ChatGPT browser extension: a setup source for the Codex or ChatGPT desktop app and its browser extension. Set up prepares the bridge components. Discovered pipes that are not yet saved appear under Found here, not saved yet. Tick one to save it and let the Agent use it.: checking one saves that connection, allows the agent to use it, and leaves the current default unchanged. These are unverified endpoint candidates (desktop app tools share the same pipe-name prefix), so a newly saved pipe reports that approval is not yet confirmed until a browser handshake succeeds. Saving does not open the pipe, approve the extension, register a native host, or restart the browser. Kiki handles a real permission request when the app or extension sends one. Official instructions stay on the route. If no candidates are listed, check the official host and runtime diagnostics; an empty list alone does not mean the extension is disabled. If saving fails, the checkbox stays clear and the page says why, so you can try again.
• Ego Lite: install Ego Lite yourself from its official download page and finish its own setup window. Kiki does not install Ego Lite and does not import Chrome data. There is no component-install button. When Ego Lite is not installed, the row links to the download. If it is already installed and its own setup is finished, this page checks that installation and can show it ready; an unfinished Ego setup window stays unfinished, and a process that merely exits is not treated as ready. Check Ego Lite runs that check again. When ready with no saved connection yet, checking its discovered item under Found here, not saved yet. Tick one to save it and let the Agent use it. saves and allows it. Advanced fields on the connection are the CLI path, an optional profile id, and an optional service name.

Each source has its own switch. Turn a source off to stop discovery, connection and tool use for that source. The connections you had checked stay checked, and turning the source back on restores those same allowed connections. A connection that was not checked, including one found while the source was off, stays unchecked.

Each source row states its own status and what it still needs, and puts the next action on that same line: Set up while a component is missing, the store link while only the approval is left, and Allow browser control while only the capability switch is. Once everything Kiki can install is in place, the Kimi and Independent rows stop asking you to install. Step details sit in the What this server checked fold.

Allow browser control is the same consented action, on this page. When the capability switch is the only thing missing, that one confirmation installs whatever is still missing — nothing at all when the components are already in place — enables the plugin, and allows browser control (launching or attaching) for the server. The confirmation says all three before you accept. Browser control is available by default; components still need consented setup and a connection starts only when requested. The action also restores a saved off choice when you explicitly confirm it.

If something outside this page holds browser control off — an environment variable, or a runtime override on that host — no action here can change it. The row says exactly that instead of offering a button, and See what holds it off names the override responsible.

Opening a row in Browser connections replaces the list with that connection's detail view, with a back button to return to the list and a read-only Allowed / Disabled tag (permissions are toggled by the checkbox in the list). Existing connections retain their id, type and default after upgrades; a display name never selects a provider. Saved Codex connections can be renamed here without changing their runtime or browser binding. Under + Add browser → Configure manually, you can create a connection in one of three styles:

• Profile (Independent agent browser): Kiki starts and manages a browser of its own on the server, with a profile (sign-in state) of its own — your everyday browser cookies are not copied. Optional extras: an installed Chrome or Edge instead of the bundled one, and Show the browser window (off by default, so it starts headless).
• CDP (Existing or remote browser): attaches to a browser that already exposes a CDP address and starts nothing new. That address names the machine the browser runs on, which may be a different machine.
• Kimi WebBridge: creates a record bound to the local Kimi WebBridge HTTP daemon on the server, with no endpoint address required.

Each connection can be checked, connected and disconnected on its own. A check starts the managed driver and reads its session and tab state; a CDP connection handshakes with the endpoint and lists its targets; a Kimi WebBridge connection checks the local bridge. None of them opens a page. Saving writes configuration only and ends that connection's current running session first. Disconnect releases the connection: a borrowed CDP or WebBridge browser stays, an instance Kiki started is closed. Default connection applies to sessions created afterwards; turning a connection off never swaps in another one.

The driver, and any browser Kiki starts, run on the Kiki server — not on the device showing this window; the components live under <Kiki home>/browser/resources on the machine that runs them.

Agents use these connections through the browser tools, naming a connection by its id — never by display name. The capability flag and the connections themselves live under [experimental] and [browser_control] respectively.

Plugins

Settings → Capabilities → Extensions & tools (/settings/extensions) has three tabs: Plugins (the default, /settings/plugins), Skills (/settings/skills), and MCP (/settings/mcp). The Plugins tab lists installed plugins. Each row carries the plugin's own name, where it came from, its version, whether it is on or reporting errors, and a switch to turn it off without leaving the list. Find plugins in the market at the top is where new plugins are installed.

Selecting a plugin opens its own settings page (/settings/plugins?plugin=<id>): the plugin's name and state, a way back to the list, its settings, and a switch to turn it on or off. An enabled plugin can supply its own settings panel. The panel can follow this app's colors and language. A panel that does not keeps the wording it shipped with. Otherwise Kiki shows the fields its manifest declares. Fields that name a group sit together under that group; a secret is marked as stored and is not shown again. Save and Discard appear after a field changes. A failed save keeps the draft and the last stored values. If the plugin's interface fails to load, choose Retry or use those fields. Leaving with unsaved changes asks first.

Remove plugin is at the bottom of the page; it asks first and names what it takes with it, as the plugin manager does. A plugin with neither settings fields nor its own settings interface still gets a page with its switch and Remove plugin.

The page has its own address, so you can copy it, refresh, or use the browser's back button and come back to the same plugin. The plugin's Plugin settings entry on the Extensions & tools page leads to it, and so does the package button on a media source.

Workspaces

Settings → Sessions & work → Workspaces (/settings/workspaces) lists the folders known to this server. Open a row to configure that workspace on its own page (/settings/workspaces?workspace=<id>): rename it, pin it, choose its plugin defaults, or set memory to inherit, on or off. Its root path is shown for identification, not edited here.

Workspace trust controls whether Kiki loads that project's configuration, MCP servers and hooks, and permits its project profiles to name external executors. Trust changes only when you explicitly use the switch; opening the page grants nothing and does not change tool approval mode. The Agents, MCP servers and Skills links keep this workspace as their target. Profiles retain their file source and write scope. MCP entries from plugins or project files are read-only here; editable user entries still save to the user configuration, not a new workspace copy. A missing workspace is reported rather than silently switching to another one.

Computer control

Settings → Capabilities → Browser & computer, then the Computer control tab (/settings/computer-control opens that tab), brings together the overall desktop verdict, the connected machine and executor installation, computer MCP connections, and model usage preference. Desktop actions run on the machine hosting the Kiki server, through a pinned open-source executor; nothing is installed or connected automatically.

Allow computer control sits at the top of this tab beside a single verdict line that states whether the agent can currently drive the desktop, which connection is active, or the first gate to resolve (such as Install executor when the active CUA path is not installed yet). It stays allowed unless this server has turned it off. On by default chooses whether a new conversation starts with the tools. Off for one conversation leaves them out of that conversation. Turning allow off leaves them out everywhere, including a task that names them. Changing allow updates the tools a computer connection that is already running offers. Tool policy can still deny those tools, or only some of them, and the page says so. The switch does not install the executor or connect the desktop. The usage preference below does not grant or remove the tools.

Install executor downloads and verifies that release, then registers a global MCP connection named kiki-computer (the driver's path plus mcp arguments). Installed on the Bundled CUA executor line means the pinned files verified — it does not check that anything can actually be controlled, and it does not touch the desktop. The machine line names the host that would be driven (Kiki server is the connected server, so over a remote or SSH connection the desktop is that host's own graphical session), with Check status and an Install details fold for step-by-step verification.

Under Computer control connections, Active connection on the list header chooses which computer MCP connection runs (defaulting to the CUA default path), and + Add manually creates a custom connection. Opening a connection row replaces the list with that connection's detail form (with a back button to return to the list) to edit its command, arguments, executor, or environment variables, Test the connection, or Stop control. On a Windows server, the Register an installed Windows-MCP 0.8.8 environment fold beneath the list registers an existing Python interpreter and --config file into the global MCP configuration without running uv, install, auth, or serve, and without automatically switching the active connection.

Model usage preference only guides the model. It does not grant or remove the permission the switch and tool policy decide. It defaults to Avoid: models generally avoid computer control unless you or your rules explicitly request it, and prefer existing CLI, API, MCP, short-script, and browser-specific capabilities. Prefer favors computer control for suitable interactive tasks. Both choices respect your current explicit instructions, tool availability, correctness, and permissions. Picking Avoid or Prefer saves immediately and applies on the next model request without changing tool permissions or installation; a failed save reverts to the stored value and reports the error. The source tag shows the effective configuration layer; Restore inherited removes this server's override and adopts the inherited value, which can also be Prefer.

The main agent of each session uses the executor's own tools through the existing MCP mechanism and permissions. Children need an explicit computer-tool opt-in; the usage preference does not enable their tools. See computer_control for the field and tool-selection rules. The pinned Windows executor watches the primary display and cannot pick a different one. On macOS the desktop permission travels in the calling process, so grant it where the driver reports it missing.

Stop control ends the cua processes this service started and disables an editable connection's configuration. Another client or another Kiki instance may still be driving that desktop.

Composer

Settings → App → General (/settings/general) holds Persist composer drafts on the Composer & session card. It controls whether new-session choices (model, effort, workspace, working directory, profile) and per-session draft text are written to this browser. Turning it off clears the saved store. See Workspace and session management.

Dispatch capabilities

On Settings → Models & agents → Agents → Agent profiles (/settings/agents), open a profile and use Child agents to decide what that agent may start. The Subagents & dispatch tab (/settings/agents?tab=dispatch) sets what applies across profiles: the default subagent, subagent model governance, the subagent panel's open mode, and the subagent tool settings. Allow creating child agents is Not declared, Allowed, or Off. Not declared follows the layer above this profile. Off stops this agent from starting child agents, including a definition supplied by path. Preset list chooses the named presets and routes it may start; a grey value is the child's own until you set one here. Recommended first is only an order, and does not allow or forbid a child. Never dispatch refuses those names even when the preset list would allow them. A missing model or preset is reported on the profile, separately from whether creating child agents is allowed. See Agents and subagents.

Session controls

Session controls stay at /settings/sessions (Sessions & work → Sessions). Sessions & work → Workspaces opens workspaces and does not replace these addresses.

That page also carries Task and background policy — how many background tasks may run, the timeout that applies to Bash tasks, and what a non-interactive run does with work still in flight — and Cron operations, a read-only readout of the scheduled-task runtime. The task fields are stored under background.

Import history (/settings/sessions/import) opens Kiki's built-in importer, also available from New session. No plugin installation or activation is needed for the built-in formats; see Session history import.

Manage archived conversations on /settings/sessions opens /archived for browsing, searching and reading archived conversations. It also offers permanent deletion of one conversation or all archives in the current connection's home, with confirmation and per-item failure recovery. See Archive management for deletion scope.

Session titles on /settings/sessions also edits the instruction used to write automatic titles. The card names whether the built-in or custom instruction is active; Show reveals the server's built-in text as a read-only reference. Edit Your instruction and save, or choose Restore the built-in default to remove your override. Saving an empty instruction also restores the default. A failed save keeps your draft. These actions preserve the title model, selected moments and automatic-title switch, affect the next title request, and leave existing titles unchanged. See session_title.

Effective prompts

In a session, open the header's ⋯ → Effective prompts. The drawer opens independently of the right rail, including on narrow screens, and shows the current agent identity, profile, model and executor, the prompt channels and their source order, field overrides with the reason for each, file locations, and the cognition-anchor scope.

The binding and disk revisions help you spot which source files changed. To actually reload the session's prompt sources, use Rebuild context once the session is idle. Check all branches is a separate action that inspects the configured common, main-agent and independent-agent prompt files without switching the current identity.

Cockpit

On desktop-width screens, Standard / Cockpit in the right-panel header switches the panel between its normal width and a wide one that takes over the preview space. The conversation and composer stay in the main column either way, and Standard or Exit cockpit puts the previous preview content, tabs, draft and width back. Opening a file, an agent detail or a skill preview also leaves cockpit.

Usage

Open Usage in the sidebar (/usage). The tabs are Live, History, Quotas, and External sync, in that order. With no history filter and no panel= choice, the page opens Live. An address that already filters history, or panel=history, opens History. panel=realtime and panel=limits open Live.

• Live: model requests and live agents appear as two vertically arranged groups, each with its own expandable details. Request details show running and queued native-request counts across this service, breakdowns by model, provider and role, and waiting rows with blocking rule IDs and elapsed queue time. Agent details show main, subagent and independent execution counts, with breakdowns by executor, profile, model or role. If the connection fails, the last counts are marked stale. On the same tab, Concurrency limits lets you add or edit rules, choose a model or provider target, and set a shared or per-session cap. The switch pauses a rule without deleting it. Saving applies the rule to new and queued requests without stopping active streams.

• History: costs and token usage for a date range, defaulting to today. Follow a trend bar to its consumption sources, then open session records and recorded turns on demand; filters stay with the view. Previous-period comparisons are read only when requested. Token usage and cost have separate completeness indicators; Data reliability distinguishes unknown accounting from empty ranges. Existing filtered usage links open History. See Usage statistics.

• Quotas: balances, credits, usage windows and other metrics reported by configured providers and services, separate from Kiki's local usage accounting. Cards show the source account, status and available meters. A source that requires authentication offers the appropriate sign-in or settings action. Signing in rereads the account and locally cached quota status without querying the vendor or turning on a paused check. Use Refresh quotas to refresh eligible sources, or a card's refresh action for one source. Missing or unsupported metrics are not treated as a zero balance.

• External sync: destinations that receive this server's own usage. Three kinds are available: vibecafe.ai, Kiki webhook, and Script. Only the model, the UTC half-hour, the four token counts, quality and cost go out — never a prompt, answer, title, workspace name or path.

  A vibecafe.ai destination signs in from the page itself: press Sign in to VibeCafe and approve the displayed code in your browser. No address, client id, key or storage choice is needed. Kiki saves the credential on the connected server, using its system keyring or, if unavailable, a private file protected by filesystem permissions rather than encryption at rest. The saved connection survives a restart. If saving fails, use Try again; the previous credential is kept. To supply your own endpoint or key, choose Kiki webhook instead.

A Kiki webhook destination takes your own HTTPS endpoint, with none, bearer, or hmac authentication and optional gzip. A secret is stored either in the system keyring or, if you choose that explicitly, in a private file on the server — the keyring is not silently substituted if it fails, and the file is protected by filesystem permissions rather than encrypted at rest.

Existing /usage?panel=limits links open Live and focus the concurrency-rule section; there is no separate Limits tab.

External sync needs no experimental switch; without a destination nothing is sent. Signing in only stores an approved credential: the destination stays disabled and no consent has been given. Choose a destination, preview the exact payload, agree once, then enable.

A destination that has not been agreed to sends nothing. Widening the range, changing the endpoint, or changing a credential to a different identity asks again, while shrinking the range or changing the interval does not. Pause, and Disable sync on a vibecafe.ai destination, turn sending off (enabled false, state disabled, no next time) while keeping the queue and signed-in account. Removing asks whether to discard queued batches and deletes the stored credential; asking the service to delete what it already holds is a separate action.

On upgrade, an explicit off choice for the retired usage_export flag pauses existing destinations once, keeping their consent and queued data; resume each destination when you want it to send again. See kiki usage-export for the command-side equivalent.

A Script destination runs your command as your own OS user with your ordinary permissions — it can read files and reach the network on its own, and this is not a sandbox. Kiki writes only the content-free batch to its stdin and reads a receipt from its output.

For a request that looks stuck, open Live → Request details before changing a limit. A queued row names the local rule blocking it, and a provider HTTP 429 is diagnosed from the provider's own error — a full local queue, a timeout and a rejection each need a different fix. request_governance has the fields, examples and error codes. The rules cover the model requests this service sends itself: when you run a turn on Codex, Claude Code, or Grok Build as the engine, those requests are the engine's own and are not counted here, while an external tool calling back into Kiki for a native request is.

CLI counterpart

The TUI has no Settings dialog. Terminal-side preferences (theme, editor, and the rest) are configured in tui.toml or with the interactive /config, /theme and /editor commands — see tui.toml. Agent and runtime settings live in config.toml.